Foundation-model evaluation harness for patient-derived tumor organoid (PDTO) drug-response prediction. Realizing the benefits of foundation models requires careful evaluations that map the boundaries of generalization.
The harness produces a registry-backed report comparing models against three out-of-distribution split strategies on PDTO datasets, with negative/positive controls, bootstrap confidence intervals, and a pretraining-leakage exposure profile per result.
# Install uv (Python package manager) if you don't already have it
curl -LsSf https://astral.sh/uv/install.sh | sh
# Sync dependencies (creates .venv and uv.lock)
uv sync --extra dev
# Run the tests
uv run pytest- Soragni 2024 sarcoma PDTOs (Synapse PDTOSarcoma)
BSD-2-Clause Plus Patent License (see LICENSE).